0G Introducing ERC-7857

The 0G Labs team has recently introduced ERC-7857, a groundbreaking NFT standard tailored specifically for AI agents that brings Intelligent NFTs (iNFTs) to life. 

iNFTs represent AI Agents as NFTs, providing numerous advantages such as transferability, decentralization, complete asset control, royalties, and more. While traditional fungible AI agents are a good first step for the space, iNFTs are the future and present a major leap forward by embedding unique metadata within the NFT framework.

As the AI agent space will continue to grow at an exponential rate, traditional NFT standards like ERC-721 and ERC-1155 have several limitations that make them ill-suited for AI agents. 

ERC-7857 provides a key solution that addresses gaps in how these agents are owned, traded, and used in Web3. Moreover, ERC-7857 directly complements 0G’s decentralized AI Operating System (deAIOS), which includes 0G Storage, 0G Data Availability (DA), 0G Services Marketplace, and more.

In this article, we will explore:

• The concept of Intelligent NFTs (iNFTs) and how they redefine AI agents.

• The issues with existing NFT standards and why they fall short for AI agents.

• How ERC-7857 works in practice, from metadata encryption to secure transfer.

• Key applications of ERC-7857, including agent marketplaces, AI-as-a-Service, and intellectual property monetization.

• Why this matters for Web3 and AI developers, and how to get started in building iNFTs with ERC-7857.

Intelligent NFTs (iNFTs)

AI Agents grew exponentially throughout 2024 by all metric, and are still only scratching the surface of what’s to come. 

That being said, the future of AI agents lies in Intelligent NFTs, known as iNFTs.

These represent the next era of crypto AI and are a fusion of Artificial Intelligence (AI) and NFTs. At present, on-chain AI agents are controlled by centralized teams that control and monetize the agents. 

With iNFTs, AI agents are represented as NFTs which introduces several key advantages:

• Transferability: AI agents can be freely bought, sold, or transferred between owners, with their value reflecting their capabilities and earnings.

• Full Decentralization: No longer owned and managed by a single team, iNFTs ensure decentralized control over AI agents.

• Asset Control: Owners have exclusive rights to manage and claim all assets, earnings, and gains generated by their AI agents.

• Royalties: Those creating AI-NFTs can earn a percentage of revenue every time their AI agents are resold on secondary markets.

Crucial to the above is that each Agent has its own metadata, which can be thought of as the agent’s “intelligence”. This data must, in turn, be securely encrypted, as it’s what differentiates each agent.

At present, existing NFT standards do not support this, which is why ERC-7857 is significant.

Issues with Existing Standards

Representing AI agents as NFTs provides a means of tokenizing their ownership and capabilities, making them fully composable within on-chain ecosystems. However, existing NFT standards fall short in addressing the specific requirements of AI agents, particularly around metadata privacy, secure transfers, and dynamic functionality.

NFT standards have introduced golden ages of various on-chain assets, with ERC-721 with ERC-721 and ERC-1155 being two of the most common standards used today. 

ERC-721 is widely adopted for representing unique, non-fungible assets such as digital art, collectibles, and in-game items. Each token is one-of-a-kind and tied to static metadata, such as a URI that points to a JSON file. ERC-1155, on the other hand, is a Multi-Token Standard that supports both fungible and non-fungible tokens within the same smart contract. It’s commonly used in gaming, where items may be fungible (e.g., potions or currency) or non-fungible (e.g., rare swords or skins). This standard also allows for batch operations, making it efficient for managing large collections of tokens.

While these standards excel in their respective domains, they are insufficient for AI agents, which introduce unique challenges:

1. Metadata is Static and Public

With ERC-721, metadata is typically stored as a URI, which points to a JSON file on IPFS or in a centralized service. This data is static and public, while AI agents have rapidly changing information as they update their models using real-time data.

This is critical as metadata is essentially an agent’s intelligence and personality, as it contains all the unique information and features that make up an agent.

2. No Secure Metadata Transfer

When ERC-721 tokens are transferred, only the ownership of the TokenID is transferred, rather than the underlying metadata. The recipient gains the NFT but doesn’t automatically gain access to any associated private or encrypted data. 

With AI agents, the buyer must also transfer access to the agent's core capabilities, such as its neural model, memory, or behavior, as well as its metadata, which requires encryption and private handling. ERC-721 cannot handle this.

3. No Native Support for Encryption or Privacy

ERC-721 doesn’t natively support encryption or private storage for sensitive metadata. This is problematic for AI agents, as many include proprietary or sensitive data (e.g., a custom-trained neural network or private user data) so they cannot be represented as iNFTs.

To illustrate the above three challenges, let’s use an example.

Imagine you own an iNFT on an ERC-721 platform. The agent is highly valuable because it has been trained on proprietary datasets and has learned to interact in a specific way. Now:

1. You sell the NFT on a marketplace.

2. The buyer receives the ERC-721 tokenId but has no access to the iNFT’s metadata because it’s stored privately.

3. Even if the metadata were accessible, it’s unclear how to transfer it securely and verifiably to the buyer while protecting its privacy.

The result is that the buyer receives an incomplete agent and cannot use it as intended, which is a major limitation to the advancement of crypto AI.

Key Features of ERC-7857

Here, we’ll provide a brief overview of the advantages of ERC-7857, before providing a walkthrough of how it works in practice.

ERC-7857 addresses the critical gaps in existing NFT standards, making it uniquely suited for representing and transferring AI agents in decentralized ecosystems. Here’s a brief overview of its advantages:

• Privacy-Preserving Metadata: Unlike ERC-721 and ERC-1155, ERC-7857 enables the secure storage and transfer of sensitive metadata. iNFT metadata, which can be considered its “intelligence” and is what differentiates them, can be encrypted and kept private, ensuring proprietary data remains protected.

• Secure Metadata Transfers: ERC-7857 facilitates the transfer of both ownership and metadata in a privacy-preserving and verifiable manner. This ensures the new owner receives the complete AI agent, including its functional capabilities, without risking data exposure.

• Fast Data Management: AI agents quickly evolve over time, with continuously updating models and behaviors. ERC-7857 supports dynamic metadata, enabling the agent’s state to be updated and preserved securely within the NFT framework.

• Integration with Decentralized Storage: By integrating with decentralized storage systems like 0G Storage, ERC-7857 ensures that metadata is permanently and securely accessible. This allows for tamper-proof data management while preserving accessibility for authorized owners.

• Verifiable Ownership and Control: ERC-7857 introduces cryptographic proofs and oracles to validate metadata transfers. This provides a layer of trust and ensures the integrity of both the metadata and the transfer process.

• AI-Specific Use Cases: ERC-7857 is purpose-built for AI agents, enabling features like agent lifecycle management. Agents can verify ownership before performing tasks, preventing unauthorized usage and enhancing functionality in marketplaces, enterprise solutions, and beyond.

By addressing the limitations of existing standards, ERC-7857 provides a robust, scalable framework for AI agents to thrive in decentralized ecosystems. Let’s now dive into the technical details of how this standard works in practice.How ERC-7857 Works in Practice

There are a few distinct steps to how this works.

Step 1: Metadata Encryption and Hash Commitment

Each AI agent's metadata (e.g., neural network model, memory, character traits) is stored in an encrypted form for privacy. A hash commitment is generated from this encrypted metadata, which serves as proof of its authenticity without exposing the actual content. 

The sender (current owner of the agent NFT) has access to the encrypted metadata and the corresponding hash.

Step 2: AI Agent Transfer

When the agent NFT is sold or transferred, the metadata must also change hands securely. To facilitate this, a trusted oracle decrypts the original metadata securely. It might use a Trusted Execution Environment (TEE), a privacy-preserving technique, to keep data private.

The oracle generates a new encryption key for the metadata and encrypts it again using this new key. The new encrypted metadata, along with its new hash, is published to a decentralized storage system such as 0G Storage.

Step 3: Receiver-Specific Key

To ensure only the new owner can access the metadata, the new encryption key is encrypted using the public key of the receiver (the buyer or new NFT owner).

This encrypted key is sent to the sender, who provides it as part of the NFT transfer process.

The receiver (new owner) acknowledges that they can access the newly encrypted metadata by signing the metadata hash with their private key. The oracle also provides proof verifying that the new encrypted metadata was derived from the original metadata.

Together, these proofs (the receiver’s acknowledgment and the oracle’s validation) form the proof parameter required for the NFT transfer" for clarity.

Step 4: Finalizing the Transfer

The transfer() interface in the ERC-7857 smart contract is invoked with:

• The sender's proof of accessibility.

• The oracle's proof of metadata validity.

• The receiver's signature acknowledges accessibility.

• The encrypted key for the metadata (encrypted with the receiver’s public key).

The smart contract verifies all these proofs. If valid, the NFT transfer is completed and the encrypted metadata key is published to the receiver.

The receiver decrypts the metadata key using their private key.

With the key, the receiver can now access the AI agent’s encrypted metadata and fully utilize the agent.

ERC-7857 In Application

ERC-7857 provides AI developers with a powerful new standard for tokenizing, trading, and securely managing AI agents as iNFTs. The ability to securely transfer both ownership and private metadata makes it an essential tool for advancing the Web3 AI space.

Some use cases that we find interesting include:

1. Agent Marketplaces: Developers or creators can sell trained iNFTs on decentralized marketplaces (like OpenSea for agents). The ERC-7857 standard ensures that buyers receive the agent’s full intelligence (metadata) securely and verifiably.

2. Personalized Automation Agents: Individuals can own AI agents customized for personal workflows, such as automating on-chain trading, claiming airdrops, or managing digital assets on their behalf. The agents, trained with what they’ve learnt, can then be sold given their status as an NFT.

3. Enterprise AI Ownership: Enterprises can build proprietary AI agents tailored to internal needs (e.g., customer support bots, and financial advisors) and securely transfer or lease these agents to subsidiaries or clients.

4. AI-as-a-Service (AIaaS): AI agents can be tokenized and leased on subscription models.

5. Agent Collaboration and Composability: AI agents can be combined or traded between developers to enhance functionality, such as combining the metadata and proprietary algorithms of a creative AI agent and data analytics agent.

6. Intellectual Property Monetization: Tokenizing AI models as NFTs enables developers to monetize their creations while retaining control over how and when they are used.

This is only scratching the surface of what’s possible, especially as the quantity and variety of AI agents continue to expand.

Want to Build with 0G?

0G is primarily renowned for our work in bringing AI on-chain through our various solutions, such as 0G Storage, 0G DA, and 0G Services Marketplace. We’ve recently raised $325M to help us grow and are busy partnering with an extensive range of top projects across every Web3 vertical.

Our introduction of ERC-7857 illustrates another step that we have taken to advance the crypto AI space, and we consider it worthwhile for any ambitious projects and developers to reach out to discuss how this standard could support their growth.

If you are interested in learning how we can support your team, please reach out via Discord.

To stay up-to-date, follow us on Twitter.